From b706fd1628b06c216a945176a9fedfa808324803 Mon Sep 17 00:00:00 2001 From: Chaojun Zhang Date: Fri, 7 Aug 2026 13:18:25 +0800 Subject: [PATCH] [CI][XPU] Work around intermittent segfault in Intel XPU CI with VLLM_DISABLE_COMPILE_CACHE=1 (#51337) Signed-off-by: Chaojun Zhang --- .buildkite/intel_jobs/basic_correctness.yaml | 3 ++- .buildkite/intel_jobs/model_executor_intel.yaml | 3 ++- .buildkite/intel_jobs/model_runner_v2_intel.yaml | 3 ++- .buildkite/intel_jobs/models_distributed_intel.yaml | 3 ++- .buildkite/intel_jobs/samplers_intel.yaml | 3 ++- 5 files changed, 10 insertions(+), 5 deletions(-) diff --git a/.buildkite/intel_jobs/basic_correctness.yaml b/.buildkite/intel_jobs/basic_correctness.yaml index 7d9011d6dc..35a1c21075 100644 --- a/.buildkite/intel_jobs/basic_correctness.yaml +++ b/.buildkite/intel_jobs/basic_correctness.yaml @@ -18,10 +18,11 @@ steps: source_file_dependencies: - vllm/ - tests/basic_correctness/test_cumem.py + # Remove VLLM_DISABLE_COMPILE_CACHE=1 once intel/intel-xpu-backend-for-triton#7682 is fixed commands: - >- bash .buildkite/scripts/hardware_ci/run-intel-test.sh 'cd tests && export VLLM_WORKER_MULTIPROC_METHOD=spawn && - pytest -v -s basic_correctness/test_cpu_offload.py && + VLLM_DISABLE_COMPILE_CACHE=1 pytest -v -s basic_correctness/test_cpu_offload.py && pytest -v -s basic_correctness/test_mem.py::test_end_to_end' diff --git a/.buildkite/intel_jobs/model_executor_intel.yaml b/.buildkite/intel_jobs/model_executor_intel.yaml index 14a853544e..ff3e59cb31 100644 --- a/.buildkite/intel_jobs/model_executor_intel.yaml +++ b/.buildkite/intel_jobs/model_executor_intel.yaml @@ -21,6 +21,7 @@ steps: - vllm/config/model.py - vllm/model_executor - tests/model_executor + # Remove VLLM_DISABLE_COMPILE_CACHE=1 once intel/intel-xpu-backend-for-triton#7682 is fixed commands: - >- bash .buildkite/scripts/hardware_ci/run-intel-test.sh @@ -30,4 +31,4 @@ steps: export VLLM_WORKER_MULTIPROC_METHOD=spawn && export PYTHONFAULTHANDLER=1 && cd tests && - pytest -v -s model_executor -m "not slow_test" --ignore="model_executor/layers/test_rocm_unquantized_gemm.py" --deselect="tests/model_executor/model_loader/test_reload.py::test_kv_scale_reload"' + VLLM_DISABLE_COMPILE_CACHE=1 pytest -v -s model_executor -m "not slow_test" --ignore="model_executor/layers/test_rocm_unquantized_gemm.py" --deselect="tests/model_executor/model_loader/test_reload.py::test_kv_scale_reload"' diff --git a/.buildkite/intel_jobs/model_runner_v2_intel.yaml b/.buildkite/intel_jobs/model_runner_v2_intel.yaml index 85a23eaef0..8d967d45a2 100644 --- a/.buildkite/intel_jobs/model_runner_v2_intel.yaml +++ b/.buildkite/intel_jobs/model_runner_v2_intel.yaml @@ -22,13 +22,14 @@ steps: - vllm/v1/attention/ - tests/v1/engine/test_llm_engine.py - tests/v1/e2e/ + # Remove VLLM_DISABLE_COMPILE_CACHE=1 once intel/intel-xpu-backend-for-triton#7682 is fixed commands: - >- bash .buildkite/scripts/hardware_ci/run-intel-test.sh 'export VLLM_USE_V2_MODEL_RUNNER=1 && cd tests && pytest -v -s v1/engine/test_llm_engine.py -k "not test_engine_metrics" && - pytest -v -s v1/e2e/general/test_context_length.py && + VLLM_DISABLE_COMPILE_CACHE=1 pytest -v -s v1/e2e/general/test_context_length.py && ENFORCE_EAGER=1 pytest -v -s v1/e2e/general/test_async_scheduling.py -k "not ngram" && pytest -v -s entrypoints/llm/test_struct_output_generate.py -k "xgrammar and not speculative_config6 and not speculative_config7 and not speculative_config8 and not speculative_config0" && pytest -v -s v1/e2e/general/test_min_tokens.py' diff --git a/.buildkite/intel_jobs/models_distributed_intel.yaml b/.buildkite/intel_jobs/models_distributed_intel.yaml index 5f2247eeaa..6beafda8c1 100644 --- a/.buildkite/intel_jobs/models_distributed_intel.yaml +++ b/.buildkite/intel_jobs/models_distributed_intel.yaml @@ -20,8 +20,9 @@ steps: - vllm/model_executor/model_loader/sharded_state_loader.py - vllm/model_executor/models/ - tests/model_executor/model_loader/test_sharded_state_loader.py + # Remove VLLM_DISABLE_COMPILE_CACHE=1 once intel/intel-xpu-backend-for-triton#7682 is fixed commands: - >- bash .buildkite/scripts/hardware_ci/run-intel-test.sh 'cd tests && - pytest -v -s model_executor/model_loader/test_sharded_state_loader.py -m "not slow_test"' + VLLM_DISABLE_COMPILE_CACHE=1 pytest -v -s model_executor/model_loader/test_sharded_state_loader.py -m "not slow_test"' diff --git a/.buildkite/intel_jobs/samplers_intel.yaml b/.buildkite/intel_jobs/samplers_intel.yaml index 41ec3c4df7..ffc7fec631 100644 --- a/.buildkite/intel_jobs/samplers_intel.yaml +++ b/.buildkite/intel_jobs/samplers_intel.yaml @@ -22,8 +22,9 @@ steps: - tests/samplers - tests/conftest.py - vllm/entrypoints/generate/beam_search + # Remove VLLM_DISABLE_COMPILE_CACHE=1 once intel/intel-xpu-backend-for-triton#7682 is fixed commands: - >- bash .buildkite/scripts/hardware_ci/run-intel-test.sh 'cd tests && - VLLM_XPU_USE_SAMPLER_KERNEL=1 pytest -v -s samplers' + VLLM_XPU_USE_SAMPLER_KERNEL=1 VLLM_DISABLE_COMPILE_CACHE=1 pytest -v -s samplers'